Estimation of Neuronal Signaling Model Parameters using Deterministic and Stochastic in Silico Training Data: Evaluation of Four Parameter Estimation Methods

This study evaluates parameter estimation methodology in the context of neuronal signaling networks. Based on the results of a previous study, four parameter estimation methods, Evolutionary Programming, Genetic Algorithm, Multistart, and Levenberg-Marquardt, are selected. All the reaction rate constants of the test case, the protein kinase C (PKC) pathway model, are estimated using the selected four methods. The estimations are done with both error and disturbance free training data from deterministic 1/1 silica simulations and with more realistic training data from stochastic in silica simulations. The results show that in overall the evolution based algorithms perform well. However, there is a clear need for further development, especially when utilizing more realistic training data.